app.py

import streamlit as st
from datasets import load_dataset
from transformers import pipeline, DistilBertForSequenceClassification, DistilBertTokenizerFast, AutoModelForSequenceClassification, AutoTokenizer, TFAutoModelForSequenceClassification

# Options for models from transformers library
MODEL_OPTS = ['finetuned', 'default', 'bertweet-base-sentiment-analysis', 'twitter-roberta-base', 'distilRoberta-financial-sentiment']
FINETUNED_OPT = MODEL_OPTS[0]
DEFAULT_OPT = MODEL_OPTS[1]

# Helper function
def map_decision_to_string(example):
    return {'decision': decision_to_str[example['decision']]}

def load_abstracts():
    dataset_dict = load_dataset('HUPD/hupd',
        name='sample',
        data_files="https://huggingface.co/datasets/HUPD/hupd/blob/main/hupd_metadata_2022-02-22.feather", 
        icpr_label=None,
        train_filing_start_date='2016-01-01',
        train_filing_end_date='2016-01-31',
        val_filing_start_date='2016-01-01',
        val_filing_end_date='2016-01-01',
    )
    abstracts = dataset_dict['train']['abstract']
    dataset_dict = []
    return abstracts

# returns loaded model and tokenizer, if any
def load_model(opt):
    if opt not in MODEL_OPTS: print("Incorrect model selection. Try again!")
    model, tokenizer = None, None

    # Load the chosen sentiment analysis model from transformers
    if opt == FINETUNED_OPT:
        tokenizer = DistilBertTokenizerFast.from_pretrained('distilbert-base-uncased')
        model = DistilBertForSequenceClassification.from_pretrained('saccharinedreams/finetuned-distilbert-base-uncased-for-hupd')
    elif opt == DEFAULT_OPT:
        return model, tokenizer
    elif opt == 'bertweet-base-sentiment-analysis':
        tokenizer = AutoTokenizer.from_pretrained("finiteautomata/bertweet-base-sentiment-analysis")
        model = AutoModelForSequenceClassification.from_pretrained("finiteautomata/bertweet-base-sentiment-analysis")
    elif opt == 'twitter-roberta-base-sentiment':
        tokenizer = AutoTokenizer.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
        model = AutoModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
    elif opt == 'distilRoberta-financial-sentiment':
        tokenizer = AutoTokenizer.from_pretrained("mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis")
        model = AutoModelForSequenceClassification.from_pretrained("mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis")
    
    elif not model and not tokenizer: 
        print("Model not loaded correctly. Try again!")

    return model, tokenizer

def sentiment_analysis(model, tokenizer):
    if model and tokenizer:
        return pipeline('text-classification', model=model, tokenizer=tokenizer)
    else: return pipeline('text-classification')

# Title the Streamlit app 'Finetuned Harvard USPTO Patent Dataset (using DistilBert-Base-Uncased)'
st.title('Finetuned Sentiment Analysis for US Patents')
st.markdown('Link to the app - [sentiment-analysis-app](https://huggingface.co/spaces/saccharinedreams/sentiment-analysis-app)')
st.markdown('Link to the model - [model repo](https://huggingface.co/saccharinedreams/finetuned-distilbert-base-uncased-for-hupd')
st.markdown('This model was finetuned on the Harvard USPTO Patent Dataset and uses Distilbert-Base-Uncased.')

abstracts = load_abstracts()
print(len(abstracts))
print(abstracts[0])
dropdown_abstracts = st.selectbox('Select one of the following abstracts from the HUPD dataset:', abstracts, index=abstracts.index(abstracts[0]))
model, tokenizer = load_model('finetuned')

# Take in user input
#user_text = st.text_input('Input text to perform sentiment analysis on here.', 'I love AI!')

# The user can interact with a dropdown menu to choose a sentiment analysis model.
#dropdown_value = st.selectbox('Select one of the following sentiment analysis models', MODEL_OPTS, index=MODEL_OPTS.index(DEFAULT_OPT))
#model, tokenizer = load_model(dropdown_value)


# Perform sentiment analysis on the user's input
result = sentiment_analysis(model, tokenizer)(dropdown_abstracts)

# Display the sentiment analysis results
st.markdown('Labels 0, 1: Not accepted, Accepted')
st.write('Sentiment:', result[0]['label'], '; Score:', result[0]['score'])